HARNESS · BUNDLE EVIDENCE
orders/checkout-flow
PARITYparity 99.987% ≥ threshold 99.5%SELF-CHECK ✓
SMALLEST COUNTEREXAMPLE
No counterexample — self-check passed.
PARITY verdicts show the last replayed fixtures instead:
fx-4196PASS
fx-4197PASS
fx-4198PASS
fx-4199PASS
fx-4200PASS
NORMALIZATION LEDGER — EVERY EXCLUDED FIELD + WHY
X-Request-Id (request header)regenerated on every replay attempt — correlates a call to its trace, carries no business meaning
traceparent / X-B3-TraceId (request + response headers)APM trace-propagation ids, unique per run by design — comparing them would fail every replay
Set-Cookie (response header)candidate mints a fresh session token on every replay; the value is opaque and non-deterministic on both stacks
response.headers.DateHTTP wall-clock stamp — always "now" at response time, legacy and candidate run milliseconds apart even in lockstep replay
response.body.processedAtapplication-level timestamp, same reason as Date — excluded everywhere it appears in a payload
response.body.requestDurationMsreplay-infra timing artifact, not application logic — comparing it would fail on hardware speed alone
REPLAY RECIPE — the exact commands
$ harness lab up orders-legacy
READY · pinned rails 3.2 image · booted 42s
$ harness record orders checkout-flow
captured 4,200 fixtures + 12,483 mbus events → fx-set-2026-07-18
$ harness verify orders checkout-flow
PARITY · 99.987% ≥ threshold 99.5% · self-check ✓ · exit 0
$ harness selfcheck orders checkout-flow
self-check ✓ · record → replay(legacy) → PARITY
SELF-CHECK ✓ — record → replay(legacy) → PARITY. This run's verdict is trusted, independent of orders/checkout-flow's own parity outcome.